Scenic spot dynamic recommendation algorithm and device based on theme diversity
阅读说明：本技术 基于主题多样性的旅游景点动态推荐算法及装置 (Scenic spot dynamic recommendation algorithm and device based on theme diversity ) 是由 黄超 曹畅 李漪阳 黄洁 苏宇 于 2020-10-29 设计创作，主要内容包括：本发明属于旅游信息服务技术领域,具体涉及基于主题多样性的旅游景点动态推荐方法及装置。本发明公开了基于主题多样性的旅游景点动态推荐方法,包括以下步骤,S101：收集游客基础信息及扩展信息,S102：收集景点的基础信息,S103：度量游客自身偏好变化,S104：度量考虑社交关系及潮流趋势导致的游客偏好变化,S105：游客预测评分计算,S106：正式推荐景点生成,S107：评估游客预测评分的综合性能；本发明能够面向游客实现考虑多样性的动态个性化旅游景点推荐,进而提升旅游景点的推荐效果；本方法在进行景点推荐时具有最高的精度,因而具有优良的性能。(The invention belongs to the technical field of tourist information services, and particularly relates to a method and a device for dynamically recommending tourist attractions based on theme diversity. The invention discloses a scenic spot dynamic recommendation method based on subject diversity, which comprises the following steps of S101: collecting basic information and extended information of the tourists, S102: collecting basic information of the scenic spots, S103: measuring the change of the preference of the tourist, S104: the measurement considers the social relationship and the change of the guest preference caused by trend, S105: calculating the predicted score of the tourist, S106: generating a formal recommended sight, S107: evaluating the comprehensive performance of the tourist prediction scores; according to the invention, the dynamic personalized tourist attraction recommendation considering diversity can be realized for tourists, so that the recommendation effect of the tourist attraction is improved; the method has the highest precision when the scenic spot recommendation is carried out, so that the method has excellent performance.)
1. The scenic spot dynamic recommendation method based on the theme diversity is characterized by comprising the following steps of: comprises the following steps of (a) carrying out,
s101: collecting basic information and extended information of tourists, and collecting the basic information and the extended information of the tourists from a tourism portal website and a social network site;
s102: collecting basic information of the scenic spots, and collecting the basic information of the scenic spots from a tourism portal website and a scenic spot official website;
s103: measuring the preference change of the tourist, obtaining the preference of the tourist on each theme at different moments by using an LDA model, and quantifying the preference change of the tourist through the preference difference of the two moments;
s104: the method comprises the steps that social relations and tourist preference changes caused by trend trends are measured, the social relations of the tourists have obvious influence on the preferences of the tourists, the overall trend trends of scenic spots change along with the passage of time, the preferences of the tourists are further influenced, and the preference changes of the tourists in a certain time period are calculated through the two aspects;
s105: calculating the tourist prediction scores, and analyzing the tourist prediction scores of the corresponding tourist attractions based on the tourist preference change in the S103 and S104;
s106: generating a formal recommended scenic spot, generating a formal recommended scenic spot set according to the prediction score calculated in the S105, and recommending the formal recommended scenic spot set to a target tourist in a visual form;
s107: evaluating the comprehensive performance of the tourist prediction scores, comparing the prediction scores of the scene sets recommended to the tourists with the real scores of the tourists, and if the difference of the comparison data is large, indicating that the preference of the tourists u changes; based on the current preference of the tourists and the real score of the tourists at a certain moment, the preference of the tourists can be dynamically updated by using a random gradient descent algorithm, the updated dynamic prediction score of the tourists is generated into a recommended scene set, and the recommended scene set is recommended to the target tourists again in a visual mode.
2. The method of claim 1, wherein the method comprises: in the above-mentioned step S103, the first step,
visitor u at t1Time tnThe variation of the theme preference at the time is recorded asThe calculation is shown as the formula (i),
the change of the preference degree of the tourists on each theme can be accurately measured;
in addition to theme preferences, guest u is at t1Time tnSeasonal popularity preference, distance preference and tourist bias at the moment can change, the preference degree of the tourist on the scenic spots is influenced, and the seasonal popularity preference change is recorded asDistance preference change is noted asAnd the changes of the guest's prejudice are recorded asEach calculated asFormula (c), (d) is shown:
by usingTo indicate that the guest u is at t1The overall preference of the time is calculated as shown in formula (c):
suppose that K dimensions are used to describe a guest's theme preferences, guest u is at t1Preference of time of dayThen there are K +3 dimensions and guest u is at t1Time tnThe variation of the overall preference of the time is shown in formula (sixthly):
in the step S104, the first step is executed,
tourist u is predicted from the perspective of similar neighbor v preference driftTime tnPreference drift of time of day, calculation thereofThe formula is shown in the formula of (c):
in formula (c), sim (u, v) represents a similarity metric function, and T representsTime tnThe time period of the time of day,to representTime tnThe set of guests scored between moments, Neighbour (u) representsTime tnA neighbor set of guest u between moments;
over time, the overall trend of scenic spots will change, and the preference of tourists is influenced, and the prediction is carried outTime tnTime of day global power flow driftAnd (2) predicting the whole power flow drift by adopting the preference change of all the tourists, as shown by a formula (r):
in the formula (b)To representThe number of guests in the set;
in S105, calculating the predicted score of the guest, based on the analysis of the variation of the guest preference in S103 and S104, providing a formula of ninthly,
in the formula ninthly, E is a scenery spot-theme matrix, mu is the average score of the user, buAnd biRespectively represent the tourist prejudice and the scenic spot prejudice,andrespectively represent the sight spots i atSeasonal popularity and distance of the time of day.
3. The method of claim 2, wherein the method comprises: in S107, the time recommended to the target tourist spot set is shorter than the time for the tourist to feed back the real score, the distribution of the theme features of the scenic spots, the scenic spot bias and the average score do not change significantly, and the prediction score for a certain scenic spot j is determinedAnd actual scoringThe difference is large, which indicates that the preference of the tourist u changes; based on gamesCurrent preferences of guests and inThe real score of the time can be dynamically updated by using a random gradient descent algorithm to calculate the dynamic prediction score of the tourist, and the calculation formula is as follows,
in equation r, η represents the learning rate in the stochastic gradient descent, λ1Representing the regularization rate in a random gradient descent.
4. The method of claim 3 for dynamic tourist attraction recommendation based on subject diversity, wherein: eta in the formula r is 0.01 and lambda1＝0.002。
5. The method of claim 1, wherein the method comprises: in S101, the basic information includes, but is not limited to, age, gender, income, family composition, academic calendar, work nature, friend relationship, praise relationship, and concern relationship; extended information includes, but is not limited to, attraction information that the guest has visited, time information, hotel check-in information, guest reviews and ratings of attractions.
6. The method of claim 1, wherein the method comprises: in S102, the basic information of the scenic spots collected from the tourism portal website and the scenic spot official website includes, but is not limited to, basic introduction of the scenic spots, ticket price, open time, and guest receiving capacity.
7. A device of a scenic spot dynamic recommendation method based on subject diversity is characterized in that: the system comprises a data capturing module, a candidate scenery spot generating module and a performance evaluation module;
the data capturing module is used for capturing and collecting basic information of tourists, historical tourism records and scenic spot description information on the Internet by using a scenic spot dynamic recommendation method;
the candidate sight spot generation module is used for generating a candidate recommended sight spot list according to the dynamic preference and sight spot characteristics of the tourists;
the recommended scenic spot generation module is used for further considering indexes such as popularity and heat of scenic spots in the candidate scenic spot list and generating a formal recommended scenic spot set considering diversity;
the result pushing module is used for recommending the recommendation result generated by the scenic spot dynamic recommendation method to the target tourist in a visual mode; a performance evaluation module for evaluating the performance of the device,
the performance evaluation module is used for comprehensively analyzing indexes such as precision, response rate, recall rate and the like of the recommendation result of the scenic spot dynamic recommendation method and evaluating the comprehensive performance of the algorithm.
Technical Field
The invention belongs to the technical field of tourist information services, and particularly relates to a method and a device for dynamically recommending tourist attractions based on theme diversity.
Background
At present, personalized recommendation technologies for tourist attractions are widely applied, and the basic idea of the methods is to recommend tourist attractions which may be interested by the tourist through analyzing interest preferences and characteristics of the tourist attractions, such as the scenic spots which the tourist has ever gone, the scenic spots which are similar to the scenic spots which the tourist has ever gone, and the like. On the basis, some technologies are improved, for example, information such as the geographical position of the tourist and the seasonal situation of the tourism is further considered, and therefore recommendation accuracy is improved.
The existing personalized tourist attraction recommendation technology has obvious defects in the aspect of considering the preference change of tourists. Firstly, the existing method mainly focuses on describing the static preference of the tourist based on the basic characteristics of the tourist, historical travel information, situation, social contact and other data. In fact, many studies have shown that guest preferences exhibit typical time-varying characteristics, and that this dynamic variability of preferences has a significant impact on travel itinerary planning and attraction recommendations. Secondly, the selected candidate sight spot set is the basis for developing the personalized recommendation of the scenic spots, the existing research mainly meets the preference of tourists to the maximum in the aspect of candidate sight spot selection, and the diversity and the novelty of the candidate sight spots are not considered. Research has shown that recommendation results lacking diversity and novelty will have a negative impact on enterprise profitability in the long term. Finally, due to the fact that the serious data sparsity problem generally exists in the field of personalized scenic spot recommendation, the application of accurate recommendation results is difficult to obtain by the traditional dynamic recommendation methods such as sliding time windows, changing instance weights and the like, and a new method needs to be explored for dynamically and individually recommending scenic spots.
Disclosure of Invention
Aiming at the problems, the invention aims to provide a novel dynamic personalized tourist attraction recommendation algorithm based on theme diversity, which comprises the following steps: based on a potential Dirichlet distribution (LDA) model and a matrix decomposition technology, aiming at data such as scores of tourists, comment texts, scenic spot description documents and the like, establishing basic preference models of the tourists for different scenic spot themes; constructing a highly interpretable tourist preference drift prediction model from six aspects of tourist theme preference, scenic spot season popularity, tourist distance, tourist bias, tourist social relationship, scenic spot trend change and the like based on a matrix decomposition technology (MF); considering the dynamic tourist score prediction algorithm, providing a recommendation algorithm which not only considers the preference of the tourist but also ensures the diversity of recommendation results; and recommending the scenic spot list meeting the preference of the tourist to the tourist based on the sorting of the recommendation result.
In order to achieve the technical purpose, the technical scheme adopted by the invention is as follows:
the scenic spot dynamic recommendation method based on the theme diversity comprises the following steps,
s101: collecting basic information and extended information of tourists, and collecting the basic information and the extended information of the tourists from a tourism portal website and a social network site;
s102: collecting basic information of the scenic spots, and collecting the basic information of the scenic spots from a tourism portal website and a scenic spot official website;
s103: measuring the preference change of the tourist, obtaining the preference of the tourist on each theme at different moments by using an LDA model, and quantifying the preference change of the tourist through the preference difference of the two moments;
s104: the method comprises the steps that social relations and tourist preference changes caused by trend trends are measured, the social relations of the tourists have obvious influence on the preferences of the tourists, the overall trend trends of scenic spots change along with the passage of time, the preferences of the tourists are further influenced, and the preference changes of the tourists in a certain time period are calculated through the two aspects;
s105: calculating the tourist prediction scores, and analyzing the tourist prediction scores of the corresponding tourist attractions based on the tourist preference change in the S103 and S104;
s106: generating a formal recommended scenic spot, generating a formal recommended scenic spot set according to the prediction score calculated in the S105, and recommending the formal recommended scenic spot set to a target tourist in a visual form;
s107: evaluating the comprehensive performance of the tourist prediction scores, comparing the prediction scores of the scene sets recommended to the tourists with the real scores of the tourists, and if the difference of the comparison data is large, indicating that the preference of the tourists u changes; based on the current preference of the tourists and the real score of the tourists at a certain moment, the preference of the tourists can be dynamically updated by using a random gradient descent algorithm, the updated dynamic prediction score of the tourists is generated into a recommended scene set, and the recommended scene set is recommended to the target tourists again in a visual mode.
In a preferred embodiment of the present invention, in S103,
visitor u at t1Time tnThe variation of the theme preference at the time is recorded asThe calculation is shown as the formula (i),
the change of the preference degree of the tourists on each theme can be accurately measured;
in addition to theme preferences, guest u is at t1Time tnSeasonal popularity preference, distance preference and tourist bias at the moment can change, the preference degree of the tourist on the scenic spots is influenced, and the seasonal popularity preference change is recorded asDistance preference change is noted asAnd the changes of the guest's prejudice are recorded asThe respective calculation is shown as a formula (c), (d):
by usingTo indicate that the guest u is at t1The overall preference of the time is calculated as shown in formula (c):
suppose that K dimensions are used to describe a guest's theme preferences, guest u is at t1Preference of time of dayThen there are K +3 dimensions and guest u is at t1Time tnThe variation of the overall preference of the time is shown in formula (sixthly):
in the step S104, the first step is executed,
tourist u is predicted from the perspective of similar neighbor v preference driftTime tnThe calculation formula of the preference drift of the time is shown as follows:
in formula (c), sim (u, v) represents a similarity metric function, and T representsTime tnThe time period of the time of day,to representTime tnThe set of guests scored between moments, Neighbour (u) representsTime tnA neighbor set of guest u between moments;
over time, the overall trend of scenic spots will change, and the preference of tourists is influenced, and the prediction is carried outTime tnTime of day global power flow driftAnd (2) predicting the whole power flow drift by adopting the preference change of all the tourists, as shown by a formula (r):
in the formula (b)To representTourist collectionThe number of guests in;
in S105, calculating the predicted score of the guest, based on the analysis of the variation of the guest preference in S103 and S104, providing a formula of ninthly,
in the formula ninthly, E is a scenery spot-theme matrix, mu is the average score of the user, buAnd biRespectively represent the tourist prejudice and the scenic spot prejudice,andrespectively represent the sight spots i atSeasonal popularity and distance of the time of day.
As a preferred scheme of the present invention, in S107, the time recommended to the target tourist spot set is shorter than the time for the tourist to feed back the real score, the distribution of the theme features, the scenic spot bias and the average score of the scenic spots are not significantly changed, and the prediction score for a certain scenic spot j is less than the time for the tourist to feed back the real scoreAnd actual scoringThe difference is large, which indicates that the preference of the tourist u changes; based on the current preferences and presence of the guestThe real score of the time can be dynamically updated by using a random gradient descent algorithm to calculate the dynamic prediction score of the tourist, and the calculation formula is as follows,
in equation r, η represents the learning rate in the stochastic gradient descent, λ1Representing the regularization rate in a random gradient descent;
in a preferred embodiment of the present invention, η ═ 0.01 and λ in r ═ r1＝0.002。
As a preferred embodiment of the present invention, in S101, the basic information includes, but is not limited to, age, gender, income, family composition, academic calendar, work nature, friend relationship, like relationship, and concern relationship; extended information includes, but is not limited to, attraction information that the guest has visited, time information, hotel check-in information, guest reviews and ratings of attractions.
In a preferred embodiment of the present invention, in S102, the basic information of the scenic spots collected from the travel portal website and the scenic spot official website includes, but is not limited to, basic introduction of the scenic spots, price of tickets, open time, and customer receptivity.
A device of a scenic spot dynamic recommendation method based on subject diversity comprises a data capture module, a candidate scenic spot generation module and a performance evaluation module;
the data capturing module is used for capturing and collecting basic information of tourists, historical tourism records and scenic spot description information on the Internet by using a scenic spot dynamic recommendation method;
the candidate sight spot generation module is used for generating a candidate recommended sight spot list according to the dynamic preference and sight spot characteristics of the tourists;
the recommended scenic spot generation module is used for further considering indexes such as popularity and heat of scenic spots in the candidate scenic spot list and generating a formal recommended scenic spot set considering diversity;
the result pushing module is used for recommending the recommendation result generated by the scenic spot dynamic recommendation method to the target tourist in a visual mode; a performance evaluation module for evaluating the performance of the device,
the performance evaluation module is used for comprehensively analyzing indexes such as precision, response rate, recall rate and the like of the recommendation result of the scenic spot dynamic recommendation method and evaluating the comprehensive performance of the algorithm.
A computer device comprises a display device, an input device, an internal/external memory, a central processing unit and an executable computer program, wherein the internal/external memory is stored with the computer program for executing the scenic spot dynamic recommendation algorithm based on the theme diversity, and the central processing unit realizes the personalized scenic spot recommendation algorithm when executing the computer program.
The invention has the beneficial effects that:
1. according to the invention, the dynamic personalized tourist attraction recommendation considering diversity can be realized for tourists, so that the recommendation effect of the tourist attraction is improved;
2. the method has the highest precision when the scenic spot recommendation is carried out, so that the method has excellent performance.
Drawings
The invention is further illustrated by the non-limiting examples given in the accompanying drawings;
FIG. 1 MAE comparison of the method to the reference method
Detailed Description
In order that those skilled in the art can better understand the present invention, the following technical solutions are further described with reference to the accompanying drawings and examples.
A method for dynamically recommending tourist attractions based on theme diversity comprises
S101: collecting basic information and extended information of tourists, and collecting the basic information and the extended information of the tourists from a tourism portal website and a social network site; basic information includes, but is not limited to, age, gender, income, family composition, academic calendar, nature of work, friend relationship, like relationship, concern relationship; extended information includes, but is not limited to, attraction information that the guest has visited, time information, hotel check-in information, guest reviews and ratings of attractions.
S102: collecting basic information of the scenic spots, and collecting the basic information of the scenic spots from a tourism portal website and a scenic spot official website; the basic information collected from the travel portal and attraction official websites includes, but is not limited to, attraction base referrals, ticket prices, open times, guest receptions.
S103: measuring the preference change of the tourist, obtaining the preference of the tourist on each theme at different moments by means of an LDA model, and quantifying the preference change of the tourist through the preference difference of the two moments;
visitor u at t1Time tnThe variation of the theme preference at the time is recorded asThe calculation is shown as the formula (i),
the change of the preference degree of the tourists on each theme can be accurately measured;
in addition to theme preferences, guest u is at t1Time tnSeasonal popularity preference, distance preference and tourist bias at the moment can change, the preference degree of the tourist on the scenic spots is influenced, and the seasonal popularity preference change is recorded asDistance preference change is noted asAnd the changes of the guest's prejudice are recorded asThe respective calculation is shown as a formula (c), (d):
by usingTo indicate that the guest u is at t1The overall preference of the time is calculated as shown in formula (c):
suppose that K dimensions are used to describe a guest's theme preferences, guest u is at t1Preference of time of dayThen there are K +3 dimensions and guest u is at t1Time tnThe variation of the overall preference of the time is shown in formula (sixthly):
s104, measuring and considering the social relationship and the change of the tourist preference caused by trend,
the social relationship of the guest has a significant influence on his preference, and the guest u predicted from the perspective of the drift of the preference of the similar neighbor v isTime tnThe calculation formula of the preference drift of the time is shown as follows:
in the formula (c), the first and second phases,sim (u, v) represents the similarity metric function, T representsTime tnThe time period of the time of day,to representTime tnThe set of guests scored between moments, Neighbour (u) representsTime tnA neighbor set of guest u between moments;
over time, the overall trend of scenic spots will change, and the preference of tourists is influenced, and the prediction is carried outTime tnTime of day global power flow driftAnd (2) predicting the whole power flow drift by adopting the preference change of all the tourists, as shown by a formula (r):
in the formula (b)To representThe number of guests in the set;
s105, calculating the predicted score of the tourist, providing a predicted formula ninthly of the dynamic score of the tourist based on the analysis of the preference change of the tourist in S103 and S104,
in the formula ninthly, E is a scenery spot-theme matrix, mu is the average score of the user, buAnd biRespectively represent the tourist prejudice and the scenic spot prejudice,andrespectively represent the sight spots i atSeasonal popularity and distance of time;
s106: generating a formal recommended scenic spot, generating a formal recommended scenic spot set according to the prediction score calculated in the S105, and recommending the formal recommended scenic spot set to a target tourist in a visual form;
s107: evaluating the comprehensive performance of the tourist prediction scores, comparing the prediction scores of the scene sets recommended to the tourists with the real scores of the tourists, and if the difference of the comparison data is large, indicating that the preference of the tourists u changes; based on the current preference and the real score of the tourist at a certain moment, the preference of the tourist can be dynamically updated by using a random gradient descent algorithm, and the updated dynamic prediction score of the tourist is generated into a recommendation scene set and is recommended to the target tourist again in a visual mode;
the time of recommending the target tourist scene set is shorter than the time of feedback of real scores by the tourists, the distribution of the theme characteristics, the scenic spot bias and the average score of the scenic spots are not obviously changed, and the prediction score of a certain scenic spot j is determinedAnd actual scoringBig differenceThe change of the preference of the tourist u is described; based on the current preferences and presence of the guestThe actual scoring at that time may be dynamically updated to guest u's preference using a motor gradient descent algorithm, with parameter updates as shown in equation (r),
in equation r, η represents the learning rate in the stochastic gradient descent, λ1Representing the regularization rate in a random gradient descent;
where eta is 0.01 and lambda1＝0.002。
In order to verify the technical effect of the scenic spot dynamic recommendation method based on the theme diversity, 112 ten thousand pieces of rating data of 542 scenic spots of 23 ten thousand tourists from 2013 to 2019 are acquired from travel and each scenic spot portal website, the method is marked as DARM, and performance comparison is performed with three typical personalized scenic spot recommendation methods, and the used three reference methods are shown in Table 1.
TABLE 1 benchmark comparison methods and profiles
The average absolute error (MAE) index is adopted, the calculation formula is as follows,
the | testset | in the above equation is the number of scores for all guests in the test data set,represents the predicted score, r, of visitor u for sight iuiIs the real score of the tourist u to the sight spot i, and the MAE measures the deductionThe lower the difference between the pre-score and the true score of the recommended model, the closer the two are, indicating that the prediction of the model is more accurate.
The prediction scores of the MF, SVD, SlopeOne and DARM models are respectively calculated and verified through a calculation formula of average absolute errors, the data results of each year are shown in figure 1, and the MAE minimum of the method can be obviously obtained through the figure 1, which shows that the method has the highest precision when the scenic spot recommendation is carried out, so that the method has excellent performance.
The embodiment provides a device of a scenic spot dynamic recommendation method based on theme diversity, which comprises a data capture module, a candidate scenic spot generation module and a performance evaluation module;
the data capturing module is used for capturing and collecting basic information of tourists, historical tourism records and scenic spot description information on the Internet by using a scenic spot dynamic recommendation method;
the candidate sight spot generation module is used for generating a candidate recommended sight spot list according to the dynamic preference and sight spot characteristics of the tourists;
the recommended scenic spot generation module is used for further considering indexes such as popularity and heat of scenic spots in the candidate scenic spot list and generating a formal recommended scenic spot set considering diversity;
the result pushing module is used for recommending the recommendation result generated by the scenic spot dynamic recommendation method to the target tourist in a visual mode; a performance evaluation module for evaluating the performance of the device,
the performance evaluation module is used for comprehensively analyzing indexes such as precision, response rate, recall rate and the like of the recommendation result of the scenic spot dynamic recommendation method and evaluating the comprehensive performance of the algorithm.
The embodiment provides a computer device, which includes a display device, an input device, an internal/external memory, a central processing unit and an executable computer program, where the internal/external memory stores a computer program for executing the above dynamic tourist spot recommendation algorithm based on theme diversity, and the central processing unit implements the above method for implementing the personalized tourist spot recommendation algorithm when executing the computer program.
The foregoing embodiments are merely illustrative of the principles of the present invention and its efficacy, and are not to be construed as limiting the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.
- 上一篇：一种医用注射器针头装配设备
- 下一篇：营销转化用户的预测方法、装置及计算机设备